← Back to Home

Tracking How Immune Cells Respond to Drugs: Parse and Nomic Build a 100-Million-Cell Dataset

The Cascade project links changes over time in single-cell RNA and secreted proteins, providing experimental evidence for AI predictions of immune responses. Data generation is underway, and whether it can become a reliable drug discovery tool still requires independent validation.

By SURL BioNews

After immune cells encounter a drug, gene activity inside the cells and the signals they release may not change in sync. To enable AI to predict how treatments affect immune responses, researchers need to connect these changes. On October 29, Parse Biosciences and Nomic Bio announced a collaboration to build Cascade, a project intended to integrate 100 million single-cell RNA sequencing profiles with more than 10,000 secreted protein profiles. Data generation is still underway. This announcement covers the collaboration and study design; the full dataset is not yet complete.

According to the design published by Nomic, the study uses peripheral blood mononuclear cells from 12 healthy donors, six male and six female, spanning at least 14 immune cell types. Each donor has more than 300 perturbation conditions at each time point, including stimuli such as cytokines, chemokines, and drugs, with responses measured at 4, 8, and 24 hours. This design aims to capture early signals and later protein secretion that measurements at a single time point could easily miss.

The two companies are responsible for different layers of measurement: Parse's GigaLab generates single-cell RNA sequencing data, while Nomic measures more than 1,000 secreted proteins. The RNA data describe gene expression across cell types; the secreted proteins provide another layer of information about the signals cells send outward. The project page lists more than 11,500 protein profiles, a more specific figure than the “over 11,000” cited in the collaboration announcement.

These measurements will be linked by donor, experimental treatment, and time point, rather than pairing each individual cell with its own protein profile. For drug discovery, researchers could use the data to test whether RNA changes in a particular cell type can predict subsequent secreted protein responses, then propose signaling pathways or candidate targets worth investigating experimentally. This is an intended use of the dataset, not a demonstrated finding.

Whether models truly learn generalizable biological patterns also requires separately designed tests. Nomic proposes holding some donors or perturbation conditions out of the training data to test models' predictive ability on unseen data. However, this announcement did not provide completed model evaluations or results demonstrating that Cascade can improve the success rate of new drug development.

Cascade will be available through data licensing and has secured commitments to participate from founding partners including GSK, Pfizer, Sanofi, Graph Therapeutics, and Turbine. This indicates demand for such experimental resources among pharmaceutical and computational drug discovery companies, but the partner list alone cannot establish data quality or model performance.

One hundred million cells can increase measurement granularity, but that does not mean they capture the biological variation of 100 million people. Whether reference data built from the blood cells of 12 healthy donors can explain responses in patients, diseased tissues, or under long-term drug treatment still requires corresponding experiments and independent validation. Cascade's value will ultimately depend on whether it helps researchers formulate more accurate immunological hypotheses that also withstand experimental testing.

References

  1. BioPharma APAC
  2. Parse Biosciences
  3. Nomic Bio
  4. Nomic Bio
  5. Genetic Engineering & Biotechnology News